The researchers, writing in Frontiers in Virtual Reality, recruited 31 participants and trained them to proficiency in a virtual reality spacecraft piloting simulation modeled on a Mars entry, descent, and landing task. The task unfolds in three sequential subtasks. First, participants survey a topographic map to select a landing site that balances scientific value against terrain hazards. Next, they manually fly the vehicle using pitch and roll commands to follow a guidance reticle while nulling wind disturbances. Finally, they control their vertical descent speed with a hand thruster, balancing a soft touchdown against a limited fuel supply. Performance across all three subtasks is scored with an integrated skill metric that ranges from 0 to 100, anchored to definitions of poor, adequate, and excellent performance.

Training began with a proficiency phase in which participants completed up to 40 trials in VR, starting at difficulty level 8 out of a possible 25. An adaptive algorithm adjusted difficulty dynamically using a 2-up, 1-down staircase rule, where each subtask changed independently based on user performance. Proficiency was defined as two consecutive excellent trials with all subtasks at difficulty level 12. Participants were then evaluated in a fixed-base physical cockpit simulator at the University of Colorado Boulder, completing five trials of static difficulty at level 12, establishing a preflight baseline between 18 and 48 hours after the proficiency session.

The experiment then replicated a spaceflight mission scenario: a five-week gap between initial training and task execution, during which skill retention was threatened. Participants were randomly assigned to one of three retention conditions. The Weekly group returned once per week for five weeks, completing ten training trials per session in VR with the adaptive difficulty algorithm. The just-in-time group returned only once, five weeks after their baseline and 18 to 48 hours before their final evaluation, for a single ten-trial adaptive session in VR. The No Training group served as a control, representing the natural skill loss that occurs when no refresher intervention is provided. All groups returned for a postflight evaluation in the physical simulator, completing five trials of static difficulty at level 12 regardless of their performance during the intervening period.

The results were striking. Final evaluation performance was significantly affected by training condition, with a Kruskal-Wallis test yielding a p-value of 0.02. Post hoc comparisons with Holm adjustment revealed that the just-in-time group achieved a median first-trial performance of 97.4, significantly higher than the No Training group’s median of 65.0, with an adjusted p-value of 0.018. The Weekly group, with a median of 79.1, fell between the two but was not significantly different from either. When the researchers calculated a skill retention metric comparing preflight and postflight performance, the just-in-time group showed a median improvement of 11.1 points, while the No Training group showed a median loss of 23.9 points, a difference that reached statistical significance at an adjusted p-value of 0.02.

Perhaps the most unexpected finding emerged from a post hoc analysis of the Weekly group. The researchers found a negative correlation between the difficulty level achieved during weekly training and final evaluation performance, with a Spearman rank correlation of -0.71 and a p-value of 0.02. In other words, participants who trained at higher difficulty levels during the weekly sessions tended to perform worse at the final evaluation. The authors suggest this may reflect overtraining or negative training, where extended practice at difficulty levels well above the performance requirements of the evaluation may have disrupted the calibrated motor responses needed for the actual task. The just-in-time group, constrained by their single session, never reached such elevated difficulty levels, which may have inadvertently protected them from this effect.

The researchers also assessed subjective training experience through surveys of flow, motivation, preparation, and confidence administered before and after each session. None of these measures were significantly impacted by retention training condition. However, the data trended in interesting directions. The just-in-time group reported the highest overall flow, perceived skill, and perceived task importance, while the No Training group reported the lowest perceived skill and the highest perceived difficulty. Notably, the Weekly group reported the lowest confidence of success for the final evaluation, lower even than participants who had received no refresher training at all, suggesting that repeated exposure to high difficulty levels may have disrupted self-efficacy rather than building it.

The implications for training design are significant, particularly for spaceflight applications. Just-in-time training paradigms are especially enticing for long-duration exploration missions, where crews must retain mission-critical skills across gaps of six to eight months between training and execution. The communication delays of a Mars mission prevent the ground-based expert facilitation used in previous just-in-time training on the International Space Station, such as on-orbit ultrasound skill maintenance. Autonomous facilitation through adaptive difficulty algorithms offers an alternative that can be initialized at any level relative to a trainee’s current skill, without relying on human operators. The results demonstrate that VR technology can create operational training scenarios with adaptive difficulty that effectively mitigate skill decay even when conducted entirely in a virtual environment.

The study has limitations that the authors acknowledge. The sample size was small, participants were primarily recruited from college populations, and the just-in-time and weekly groups differed in both total training quantity and cadence, making it difficult to disentangle these factors. The personalized adaptive algorithm also created between-subject variability in training difficulty that is challenging to account for in the results. Future work could explore machine learning algorithms for more sophisticated identification of skill gaps, investigate retention over longer durations that better match real mission timelines, and examine how physiological deconditioning during spaceflight may interact with skill decay. Still, the core finding stands: a single adaptive just-in-time session in VR was sufficient to preserve complex piloting skills across a five-week gap, outperforming both no training and more frequent weekly practice, and offering a promising path for maintaining critical skills in high-consequence industries where trainee time is finite and the cost of forgetting is severe.

Subject of Research: Adaptive just-in-time VR training for skill retention in spacecraft piloting simulations

Article Title: Adaptive just-in-time training supports skill retention in virtual reality simulations

Article References: Putman, E., Sherman, S., Bonarrigo, L., Clark, T. K., & Hayman, A. P. A. (2026). Adaptive just-in-time training supports skill retention in virtual reality simulations. Frontiers in Virtual Reality, 7, Article 1866989. https://doi.org/10.3389/frvir.2026.1866989

Image Credits: AI Generated

DOI: 10.3389/frvir.2026.1866989

Keywords: virtual reality, just-in-time training, skill retention, adaptive difficulty, spacecraft piloting, training simulation, skill decay, human factors, aerospace training, refresher training, Adaptive, just-in-time